- P50
- —
- P95
- —
- P99
- —
Runner not reporting
Hyperliquid markets
Connecting same-vantage streams
Loading the benchmarkConnecting same-vantage streams
Loading the benchmark// Live benchmark
The 1h trace follows P99 BBO. Every value runs from the Hyperliquid event timestamp to decoded, validated, canonical-book-ready content at the same NRT observer.
Runner not reporting
Runner not reporting
Runner not reporting
hyperliquid-market-benchmark-v1 is accepted. No Axiom window exists for BBO, AP Northeast, AWS, in this 1h range.// Selected-range trace
Selected trace · BBO · NRT / AWS · P99 · 1h
Runner proof · BBO · NRT / AWS
// Selected-range evidence
Evidence scope · BBO · NRT / AWS · P99 · 1h
Scoped to non-overlapping intervals inside the selected 1h. The P99 chart above remains a trace of exact rolling five-minute windows, never a re-percentiled range aggregate.
The page will not infer fastest share or stability from overlapping five-minute percentile windows. Evidence appears only after sealed 30-second outcome intervals pass the same exact three-copy contract.
// Source comparison
Latest exact five-minute scope · BBO · NRT / AWS
P50, P95, and P99 are absolute event-to-canonical-book-ready latency from one exact rolling five-minute cohort. Complete P99 values render immediately; below 1,000 samples they are visibly marked as confidence-warming, never ready.
| Source | P99 latency | P95 latency | P50 latency | Cohort role / coverage | Samples |
|---|---|---|---|---|---|
| — | — | — | Match to Foundation · — | 0 | |
| — | — | — | Reference set | 0 | |
| — | — | — | Match to Foundation · — | 0 |
// Global cash sessions
Global context over selected trace · BBO · NRT / AWS · P99 · 1h
These global session events are overlaid only against the selected NRT / AWS latency trace. UTC is derived from each exchange’s local schedule and IANA timezone for the selected date.
No active or upcoming maintenance, and no completed window in the last 24 hours. Nothing is inferred from the day of week.
Open official history ↗Normal full trading days only. A venue holiday and half-day feed must be connected before production “open now” claims are enabled.
// Methodology
The benchmark publishes absolute latency, its cohort rules, public runner identity, and the filtered data shown above. It never turns a missing row into zero or a latency delta into a provider measurement.
Each path stops at the same boundary: the book has been decoded, validated, and canonicalized at the observer. Values are milliseconds from the Hyperliquid event timestamp, not a winner-relative delta.
Foundation defines the reference event set. A sample enters a percentile only when Quicknode gRPC, Foundation, and Hydromancer normalize to the same canonical book inside one sealed three-source cohort.
The collector calculates P50, P95, and P99 over one rolling five-minute window and publishes every 30 seconds. A complete P99 cohort is visible from its first matched sample and remains labeled confidence-warming until 1,000; overlapping windows are never summed or re-percentiled.
Foundation is labeled as the reference set. Quicknode gRPC and Hydromancer coverage report how much of that reference set matched; the page never presents Foundation’s role as a competing match percentage.
Every response identifies one public runner, logical region, cloud, and physical metro. LAX remains LAX and SJC remains SJC; locations and clouds are not combined into a synthetic global result.
Cash-session UTC boundaries come from local schedules and IANA timezones. Maintenance comes only from the public Hyperliquid Statuspage API. Planned bounds and official in-progress/completed updates remain separate; an unavailable feed produces no overlay. Inspect the source ↗
Canonical Hyperliquid delivery telemetry, measured continuously across a global multi-cloud probe fleet.